Proxy endpoints - bridging clinical trials and real world data

Area of Science:

• Biomedical Informatics
• Health Data Science
• Clinical Trial Analysis

Background:

• Randomized Controlled Trials (RCTs) use specific disease severity endpoints to monitor treatment effects.
• Real-world data (RWD) often lacks these precise endpoints, limiting its use for drug efficacy evaluation in larger populations.

Purpose of the Study:

• To develop a method for learning proxy models of disease endpoints within RWD.
• To enable the utilization of RWD for evaluating drug efficacy across broader patient groups.

Main Methods:

• An ensemble technique was developed using a multi-stage learning framework on RCT data.
• Feature selection identified significant disease drivers present in RWD.
• Explainable Boosting Machines (EBMs) were used to create interpretable proxy models, capturing non-linear relationships.

Main Results:

• The approach was demonstrated on rheumatoid arthritis (RA) and atopic dermatitis (AD).
• The combined feature selection and prediction method showed strong performance for both diseases.
• The results improved upon existing methods for predictive disease severity scoring.

Conclusions:

• Accurate disease severity tracking over time is crucial for patient understanding and management.
• The framework enables new applications in RA and AD, including treatment effect estimation and prognostic scoring on RWD.
• The method is extendable to other diseases where severity is poorly captured in electronic health records.